fix handling of negative offsets when encoding disp8N(thanks to rsdubtso)

readme.md

@@ -1,5 +1,5 @@
 
-Xbyak 5.33 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+Xbyak 5.34 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
 =============
 
 Abstract
@@ -323,6 +323,7 @@ The header files under xbyak/ are independent of cybozulib.
 
 History
 -------------
+* 2016/Dec/09 ver 5.34 fix handling of negative offsets when encoding disp8N(thanks to rsdubtso)
 * 2016/Dec/08 ver 5.33 fix encoding of vpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w} for disp8N
 * 2016/Dec/01 ver 5.32 rename __xgetbv() to _xgetbv() to support clang for Visual Studio(thanks to freiro)
 * 2016/Nov/27 ver 5.31 rename AVX512_4VNNI to AVX512_4VNNIW

readme.txt

@@ -1,5 +1,5 @@
 
-    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.33
+    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.34
 
 -----------------------------------------------------------------------------
 ◎概要
@@ -333,6 +333,7 @@ cybozulibは単体テストでのみ利用されていて、xbyak/ディレク
 -----------------------------------------------------------------------------
 ◎履歴
 
+2016/12/07 ver 5.34 disp8N時の負のオフセット処理の修正(thanks to rsdubtso)
 2016/12/06 ver 5.33 disp8N時のvpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w}のバグ修正
 2016/12/01 ver 5.32 clang for Visual Studioサポートのために__xgetbv()を_xgetbv()に変更(thanks to freiro)
 2016/11/27 ver 5.31 AVX512_4VNNIをAVX512_4VNNIWに変更

test/make_512.cpp

@@ -2104,8 +2104,22 @@ public:
 			}
 		}
 		{
+			const int tbl[] = {
+				-1024, -512 -256, -128, -64, -32, -16, -8, -4, -2, -1,
+				0, 1, 2, 4, 8, 16, 32, 64, 128, 256, 512
+			};
+			for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+				char xs[128], ns[128];
+				int v = tbl[i];
+				CYBOZU_SNPRINTF(xs, sizeof(xs), "zmm0, zmm1, ptr_b[eax%+d]", v);
+				CYBOZU_SNPRINTF(ns, sizeof(ns), "zmm0, zmm1, [eax%+d]{1to16}", v);
+				put("vaddps", xs, ns);
 			}
 		}
+#ifdef XBYAK64
+		put("vfmadd231ps", "zmm8, zmm31, ptr_b[r14+rbp-0x1e4]", "zmm8, zmm31, [r14+rbp-0x1e4]{1to16}");
+#endif
+	}
 	void putAVX512()
 	{
 #ifdef MIN_TEST

xbyak/xbyak.h

@@ -105,7 +105,7 @@ namespace Xbyak {
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 4096,
-	VERSION = 0x5330 /* 0xABCD = A.BC(D) */
+	VERSION = 0x5340 /* 0xABCD = A.BC(D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
@@ -1506,8 +1506,9 @@ private:
 					mod = mod01;
 				}
 			} else {
-				uint32 t = disp / disp8N;
-				if (t * disp8N == disp && inner::IsInDisp8(t)) {
+				// disp must be casted to signed
+				uint32 t = static_cast<uint32>(static_cast<int>(disp) / disp8N);
+				if ((disp % disp8N) == 0 && inner::IsInDisp8(t)) {
 					disp = t;
 					mod = mod01;
 				}

xbyak/xbyak_mnemonic.h

@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "5.33"; }
+const char *getVersionString() const { return "5.34"; }
 void adc(const Operand& op, uint32 imm) { opRM_I(op, imm, 0x10, 2); }
 void adc(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x10); }
 void adcx(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }